Ph.D. Project: Signal Processing with Feedback Networks

The aim of this project is the analysis of feed-back neural networks, including the learning process, optimization of model structure and statistical validation.
A essential property of an adaptive system is adequate training performance. However, it is generally accepted that training feedback networks is a difficult task. The project concerns the analysis of mechanisms complicating training and suggests second order training methods.
The use of feedback networks calls for an analysis of stability and robustness. By considering the network as a dynamical system, the project objective is clarify stability issues.
Finally, the project is devoted to the study of model structure optimization. In particular, the study focuses on whether existing methods for feed-forward networks can be applied to feed-back networks as well. Further, methods for validation of model structures is under development.
The feed-back networks are primarily analyzed in connection with time-series modeling/prediction problems.